John Snow, William Farr and the 1849 Outbreak of Cholera That Affected London: a Reworking of the Data Highlights the Importance of the Water Supply

Home , Hammersmith Bridge, New River (London)

Public Health (2004) 118, 387–394

John Snow, William Farr and the 1849 outbreak of cholera that affected London: a reworking of the data highlights the importance of the water supply

P. Binghama,*, N.Q. Verlanderb, M.J. Cheala aIsle of Wight Primary Care Trust, Whitecroft, Sandy Lane, Newport, I.O.W PO30 3ED, UK bPHLS Statistics Unit, Colindale, UK

Received 11 August 2003; received in revised form 8 January 2004; accepted 24 May 2004

KEYWORDS Summary Objectives. This paper examines why Snow’s contention that cholera was John Snow; William Farr; principally spread by water was not accepted in the 1850s by the medical elite. Cholera; Logistic The consequence of rejection was that hundreds in the UK continued to die. regression Methods. Logistic regression was used to re-analyse data, ﬁrst published in 1852 by William Farr, consisting of the 1849 mortality rate from cholera and eight potential explanatory variables for the 38 registration districts of London. Results. Logistic regression does not support Farr’s original conclusion that a district’s elevation above high water was the most important explanatory variable. Elevation above high water, water supply and poor rate each have an independent signiﬁcant effect on district cholera mortality rate, but in terms of size of effect, it can be argued that water supply most strongly ‘invited’ further consideration. Conclusions. The science of epidemiology, that Farr helped to found, has continued to advance. Had logistic regression been available to Farr, its application to his 1852 data set would have changed his conclusion. Q 2004 The Royal Institute of Public Health. Published by Elsevier Ltd. All rights reserved.

Introduction Farr, the dominant epidemiologist of the day, who had concluded that the available data not only Snow’s hypothesis that cholera could be spread in supported miasma (spread via atmospheric water, food or hand-to-mouth was not accepted in vapours) but also indicated that there was an the 1850s, arguably because Snow was not a underlying ‘natural law’ linking infection with 2 member of the medical elite. Although he gave cholera inversely to elevation above high water. chloroform to Queen Victoria for pain relief in child This paper re-analyses data that were contested birth, he was not one of the medical signatories on between Snow and Farr. the birth announcement for Prince Leopold.1 Instead, the medical fraternity supported William John Snow

*Corresponding author. Tel.: þ44-1983-535409; fax: þ44- 1983-537140. Snow ﬁrst published his hypothesis on the E-mail address: [email protected] causation of cholera in 1849, shortly after

elevation was the most important: ‘The elevation Table 1 Deaths from cholera in London, registered from 23 September 1848 to 25 August 1849. of the soil in London has a more constant relation with mortality from cholera than any other known Sectors Population Deaths Death rate element’.5 Farr accepted that there was of London in 1841 from per 1000 an association between mortality from cholera cholera inhabitants and a district’s water supply, and personally West 300,711 533 1.77 exhorted the water company to make 6 North 375,971 415 1.10 improvements. However, Farr also pointed out Central 373,605 920 2.46 that if registration districts were sorted into three East 392,444 1597 4.07 ‘water supply groups’ and then stratiﬁed accord- South 502,548 4001 7.96 Total 1,945,279 7466 3.84 ing to elevation, an elevation effect was still present (see Table 4). ‘While the effects of water Snow J. On the Mode of the Communication of Cholera. of the districts are apparent, they do not, in this London: John Churchill; 1849. analysis, conceal the effects of elevation’.5

Farr’s natural law the second outbreak had swept through England.3,4 In support of his hypothesis, initially Snow only had Farr progressed his concern with elevation anecdotal evidence and the data reproduced in by sorting registration districts into ‘terraces’, Table 1. 0–19-feet elevation, 20–39-feet elevation etc. Snow pointed out that more deaths had and calculated the average mortality for each occurred in the south districts of London than all terrace. Farr noted that if the ‘observed’ mor- the other districts put together (3465 deaths, tality of the 0–19-feet terrace (102 deaths per death rate 2.4 per 1000 inhabitants).3 He 10,000 population) was divided successively by 2, suggested that this was because water for the 3, 4, 5 and 6, the observed mortality of successive south districts was taken from the River Thames terraces was approximated (see Table 5). below Vauxhall Bridge, where the river was Farr concluded that this reflected a natural law polluted by sewage discharged into the Thames and went on to derive a formula that, he higher in its course. The other districts of London demonstrated, predicted observed mortality at either abstracted from the cleaner, high reaches subdistrict level.5 of the Thames or from its tributaries. Snow obtained his data from the Weekly Return of the Registrar-General that was published by William Farr, the Compiler of Abstracts to the General Data and methods Register Office. The data re-analysed in this paper, first published 5 William Farr by Farr in 1852, relate to deaths from cholera that were registered in London during 1849 (a total In 1852, Farr published additional data on deaths of 14,137 deaths, see Table 2). This was part of that had occurred from cholera in London during the second outbreak of cholera to affect the UK 1849, including eight potential explanatory and began on 22 September 1848 when a sailor variables for the 38 registration districts of London was diagnosed with the condition in Southwark. (see Table 2).5 Farr examined the association In total, 652 residents died from cholera in between death from cholera and the explanatory London in 1848. The outbreak continued until variables by ‘grouping’ the data (see Figs. 1–4), the end of 1849. and by comparing the death rate for the 19 districts with the highest values of the explanatory Explanatory variables variable with the 19 districts with the lowest values (see Table 3). Farr’s original study included eight explanatory Farr found an association between death from variables (see Table 2), as follows. cholera and all of the eight explanatory variables. However, on the basis of a more consistent Elevation relationship with elevation (see Fig. 1)and The elevation above the Trinity high-water mark the relatively large differential between the 19 of each registration district was estimated by most-elevated and 19 least-elevated districts Captain Dawson RE of the Tithe Commission. (1:3), he concluded that the association with Clearly a ‘mean’ elevation is of questionable The water supply and the 1849 cholera outbreak in London 389

Table 2 Deaths from cholera in London, registered in 1849 by registration district together with eight possible explanatory variables.

District Deaths from Elevation Annual deaths Persons Persons per Average Annual Poor rate Water cholera in above high from all causes per acre inhabited annual value of precept supplya 1849 per water (feet) 1838–1844 house value of house per per pound 10,000 per 10,000 house (£) person (£) of house inhabitants inhabitants value

Newington 144 22 232 101 5.8 22 3.788 0.075 1 Rotherhithe 205 0 277 19 5.8 23 4.238 0.143 1 Bermondsey 161 0 264 66 6.2 18 3.077 0.134 1 St George 164 0 267 181 7.0 22 3.318 0.089 1 Southwark St Olave 181 2 281 114 7.9 35 4.559 0.079 1 St Saviour 153 2 292 141 7.1 36 5.291 0.079 1 Westminster 68 2 260 70 8.8 36 4.189 0.076 1 Lambeth 120 3 233 34 6.5 28 4.389 0.039 1 Camberwell 97 4 197 12 5.8 25 4.508 0.072 1 Greenwich 75 8 238 18 6.8 22 3.379 0.038 2 Poplar 71 10 241 15 6.2 44 7.360 0.081 2 Chelsea 46 12 287 62 7.1 29 4.210 0.060 1 Hammersmith, 33 12 228 11 6.6 33 5.070 0.067 3 Brompton, Kensington and Fulham St George in 42 15 289 195 6.9 32 4.753 0.039 2 the East Stepney 47 16 242 85 6.3 20 3.319 0.080 2 Belgrave 28 19 194 65 8.0 67 8.875 0.066 1 Wandsworth 100 22 198 4 6.2 29 4.839 0.018 1 West London 96 28 302 212 9.7 65 7.454 0.072 2 Whitechapel 64 28 290 194 8.1 26 3.388 0.067 2 Lewisham 30 28 173 2 5.8 27 4.824 0.075 2 St Martin-in- 37 35 240 81 10.3 119 11.844 0.049 2 the-Fields Bethnal Green 90 36 239 115 6.3 9 1.480 0.039 2 London city 38 38 214 129 7.1 117 17.676 0.136 2 East London 45 42 259 284 8.3 38 4.823 0.056 2 St James 16 43 212 222 10.4 128 12.669 0.088 3 Westminster Shoreditch 76 48 251 161 6.6 20 3.103 0.023 2 St Luke 34 48 276 242 7.8 28 3.731 0.082 2 Hanover Square 8 49 179 57 9.5 153 16.754 0.081 3 and Mayfair Strand 35 50 242 254 10.1 66 7.374 0.018 2 Holborn 35 53 266 235 9.7 52 5.883 0.047 2 Hackney 25 55 197 14 5.9 25 4.397 0.034 2 Clerkenwell 19 63 242 202 8.2 33 4.138 0.074 2 St Giles 53 68 269 221 11.0 60 5.635 0.057 2 Paddington 8 76 197 32 7.3 64 9.349 0.052 3 St Pancras 22 80 222 59 8.8 41 4.871 0.039 2 Islington 22 88 200 28 6.6 35 5.494 0.042 2 Marylebone 17 100 227 102 9.8 71 7.586 0.030 3 Hampstead 8 350 202 5 7.2 40 5.804 0.043 3

Table sorted by district elevation above Trinity high-water mark. a 1, Thames between Battersea Bridge and Waterloo Bridge; 2, New River/Rivers Lea and Ravensbourne; 3, Thames between Kew and Hammersmith.

validity, not least because districts varied in size for the 1866 outbreak of cholera; however, this between 136 and 17,224 acres. (Note: small paper focuses on the validity of Farr’s analysis corrections were made to the estimation of that was questioned by Snow, rather than the mean elevation in data subsequently published validity of the data itself). 390 P. Bingham et al.

Figure 1 Death from cholera vs. elevation. Figure 3 Death from cholera vs. persons per acre.

Crowding Water supply Farr included two measures of crowding: persons Farr divided the registration districts of London, per house and persons per acre. based on their water supply, into three (see Table 4): districts that obtained their water from the Thames Wealth above Hammersmith Bridge; districts that obtained Farr calculated two measures of wealth: the their water from tributaries of the Thames average value of a house within a district and (Rivers New, Lea and Ravensbourne); and districts the average house value per head of the district that obtained water from below Battersea population. Bridge. Vauxhall Bridge, suggested as a boundary by Snow, is located between Battersea Bridge and Poor rate Waterloo Bridge. The poor rate precept per pound of house value (if the precept was set at 0.1 and a person owned a Reworking of data and statistical methods house valued at £10, they paid £1 poor rate). Poor rate was determined by the number of district The outcome variable in Farr’s analysis was the residents qualifying for relief, with the total house number of deaths from cholera in 1849. In the value of the district forming the denominator. re-analysis, the outcome is assumed to follow a Farr’s tables omitted the poor rate precept for binomial distribution with the number of Hampstead. Records held at the House of individuals at risk, i.e. the population totals, Commons’s Library do not give a precise ﬁgure but as the denominators. Logistic regression has been the precept is estimated to lie between 0.045 and used to ﬁt and compare a series of models to assess 0.050 (Andrew Presland, personal communication). the form of associations between the possible explanatory variables and variation in the risk of Annual mortality death from cholera on a natural log (odds) scale. The average annual mortality from all causes per Effects are estimated as changes in log (odds), 10,000 inhabitants between 1838 and 1844. i.e. log (odds ratios).

Figure 2 Death from cholera vs. house/shop value per Figure 4 Death from cholera vs. annual mortality from person. all causes 1838–1844. The water supply and the 1849 cholera outbreak in London 391

Table 3 Comparison of the cholera death rate for the 19 Table 5 Observed mortality compared with expected for districts with the highest values of the explanatory variable ‘terraces of London’. with the 19 districts with the lowest values. Elevation of Number of Observed 102 divided by Explanatory variable Mortality from cholera per 10,000 the ‘terrace’ registration mortality 1, 2, 3, 4, 5 and population for the 19 districts above the districts from cholera 6 (‘expected’ with the highest values of the Trinity per 10,000 mortality) explanatory variable compared high-water inhabitants with the 19 districts with the mark (feet) lowest values (ratio) 0–19 16 102 102 Elevation 33:100 (1:3.0) 20–39 7 65 51 Average house value 43:90 (1:2.1) 40–59 8 34 34 per person 60–79 3 27 26 Persons per acre 71:61 (1.16:1) 80–99 2 22 20 Annual death rate 84:48 (1.75:1) 100–120 1 17 17 1838–1844 340–359 1 8 –

Results However, in this paper, the rather unusual situation exists where there is no missing data Estimated odds ratios and P values from a logistic and so the model selection strategy referred to regression model with a simple linear trend with above would not lead to an increase in the the poor rate for Hampstead set at 0.045 are number of available observations. Hence, the shown in Table 6. The results with the poor rate model as shown in Table 6 is presented, as this for Hampstead set at 0.05 are very similar. table also shows useful null information. No interactions between the explanatory variables Table 6 shows that there is a significant werefoundtobesignificantatP , 0:05: association between water supply and risk of In particular, the interaction between water death from cholera, with water from the Thames source and elevation was not significant between Battersea Bridge and Waterloo Bridge ðP ¼ 0:22Þ: A simpler model than that shown in being significantly worse than either water from Table 6 could be produced by removing the most New River/Rivers Lea and Ravensbourne or from insignificant variable with either a forwards or the Thames above Hammersmith Bridge with an estimated decrease in risk of 41 and 60%, backwards stepwise procedure. The main purpose respectively. There is also a significantly of such a procedure would be to increase the decreased risk with increasing elevation and number of observations for model estimation. decrease in the poor rate. None of the other variables appear to be associated with death from Table 4 Cholera mortality rates by registration district cholera. sorted according to their source of water supply and stratified Figs. 5 and 6 show the estimated relationships by elevation. between death rate and poor rate and death rate Source of water Mortality from Elevation of the and elevation, respectively, for the three water supply cholera per 10 ‘stratum’ above supply areas, illustrating the results given in 000 inhabitants the Trinity Table 6. The left-hand plots are on a log scale and high-water the right-hand plots are on the original scale. mark (feet) The data are plotted as squares, triangles and Thames between circles and the fitted models are shown as curves. Kew and Hammersmith In order to put the apparent effect of elevation Three highest districts 11.0 175.0 and poor rate into context, Table 7 shows the Three lowest districts 19.0 35.0 change in estimated risk over the interquartile New River/Rivers Lea range of these variables. Between district elevation and Ravensbourne Ten highest districts 36.6 59.5 of 8 and 50 ft, the risk of death from cholera Ten lowest districts 59.0 24.2 decreases by 32.7%. Between a district poor rate of £0.039 and £0.079, the risk of death from cholera Thames between Battersea Bridge and Waterloo Bridge decreases by 31.4%. Six highest districts 77.0 10.0 It is suggested that, based on the size of effect, Six lowest districts 168.0 0.3 water supply most strongly invited further consideration. 392 P. Bingham et al.

Table 6 Odds ratios, 95% conﬁdence limits (CL) and P values for logistic regression model.

Explanatory variable Low 95% CL Odds ratio High 95% CL P

Constant 6.006 £ 1024 0.002626 0.01149 – Water from Thames between Battersea Bridge and Waterloo 1.00 ,0.001 Bridgea Water from New River and Rivers Lea and Ravensbourne 0.44 0.59 0.79 Water from Thames between Kew and Hammersmith 0.22 0.40 0.72 Increase in elevation above high water (10 feet) 0.85 0.91 0.98 ,0.01 Decrease in poor rate (£/100) 0.87 0.91 0.96 ,0.001 Average annual death rate 1838–1844 1.00 1.00 1.01 0.48 Persons per inhabited house 0.89 1.03 1.19 0.71 Persons per acre 1.00 1.00 1.00 0.67 Average house value per person (£) 1.00 1.00 1.00 0.35 Average house value within district (£) 1.00 1.00 1.00 0.79

a Baseline.

Discussion suggested that water supply most strongly invited further consideration. This paper details why Farr concluded that The lack of specificity in defining the water mortality from cholera was explained by a supply to districts (more than three companies registration district’s elevation above high water: were involved) may partly explain the the consistent trend with elevation (Fig. 1); observed association between elevation and the large difference in mean mortality mortality. The explanation for the association between the 19 least-elevated districts compared with the poor rate is less clear, but it is possible with the 19 most-elevated districts (Table 3); the that mortality was very high among those claiming persistence of an elevation effect when districts poor relief. were sorted into water supply groups (Table 4); Farr’s conviction that mortality was more closely and, finally, a relationship that Farr suggested linked to elevation than water supply was highly reflected an underlying ‘natural law’ (Table 5). influential, and was used by the public health Modern logistic regression that makes best use of establishment of the day to justify the miasma all the data, however, shows that three variables theory. Even in 1855, the Scientific Committee that are independently associated with mortality from investigated the previous year’s visitation of cholera. On the basis of the size of effect, it is cholera that included the Broad Street pump

Figure 5 Estimated relationship of log (odds of death) and death rates with poor rate for three water supplies in a registration district with the mean annual death rate, housing density, elevation, area, number of houses and house value. See Fig. 6 for key. The water supply and the 1849 cholera outbreak in London 393

Figure 6 Estimated relationship of log (odds of death) and death rates with elevation for three water supplies in a registration district with the mean annual death rate, housing density, poor rate, area, number of houses and house value.

Table 7 Estimated change in the risk of cholera over the interquartile range for ‘elevation’ and ‘poor rate’.

Lower quartile Median Upper quartile Interquartile range Change in risk of cholera over interquartile range %

Elevation above high water (feet) 8 28 50 42 32.7 Poor rate (£/100) 3.9 6.7 7.9 4 31.4 outbreak concluded that “On the whole evidence, it and through airborne transmission. In addition, seems impossible to doubt that the influences, Farr hypothesized that the agent (cholera which determine in mass the geographical flux) was heavier than water and therefore distribution of cholera in London, belong less to might be at higher concentrations in water the water than to air… mortality has more nearly pipes at low elevations than in pipes at high followed the degrees of elevation of soil than been elevations.9 proportionate to any other general influence we Farr was a great statistician but application of a could measure.”7 modern technique (logistic regression) to his 1852 An earlier understanding of the predominant role cholera data set demonstrates that, in this of a contaminated water supply in cholera instance, his analysis led to the wrong conclusion transmission might well have prevented cases, being reached. particularly during the 1854 outbreak. It should, however, be noted that changes to London’s water supply were already in hand. The Water Supply Acknowledgements (Metropolis) Act of 1852 required companies to move their intakes from the River Thames by 31 August Dr Tony Swan of the PHLS Statistics Unit undertook 1856 to above the tidal flow at Teddington lock. initial work on a logistic model for the data. Andrew By 1866, when England was affected by a further Presland, House of Commons’s Library. outbreak of cholera, Farr had realized that a contaminated water supply could be a direct cause of cholera and was quick to associate a sudden increase in cases in East London with References contamination of a reservoir.8 Farr, however, continued to be impressed by 1. Mackintosh JM. Snow—the man and his times. Proc R Soc Med the influence of elevation. While he accepted 1955;48:1004—7. 2. Farr W. Influence of elevation on the fatality of cholera. J Stat that local outbreaks could be waterborne, Farr Soc Lond 1852;15:155—83. continued to believe that the ‘background rate of 3. Snow J. On the mode of the communication of cholera. cholera’ was accounted for by personal contact London: John Churchill; 1849. 394 P. Bingham et al.

4. Snow J. On the pathology and mode of the communication 7. General Board of Health. Report of the committee of cholera. Lond Med Gazette 1849;745—52. see also for scientific inquiries in relation to the cholera epidemic p. 923-9. of 1854. Her Majesty’s Stationery Office; 1855. p. 48. 5. Registrar-General. Report on the mortality of 8. Registrar-General. Report on the cholera epidemic of 1866. cholera in England 1848—49. Her Majesty’s Stationery Office; Her Majesty’s Stationery Office; 1868. 1852. 9. Eyler JM, Farr W. Willaim Farr (1807—1883): An 6. Farr, W. Weekly return of the Registrar-General of births and intellectual biography of a social pathologist. University of deaths in London. 1853;14:429—32. Wisconsin; 1971.